Two files on one record
In a fictional case, monthly CSV and yearly archive share a title. Choose the period and retain distribution identity; do not sum both versions.
In-depth practical guides
Distinguish dataset, distribution, service and catalogue record before downloading or citing public data.
Updated :

AI-generated illustration
A catalogue can describe data without hosting its files. DCAT links these objects, but portals may leave properties incomplete. Use the record to find the producer and analysed version.
Identify dataset, catalogue record, distributions and access services. A CSV and an API do not automatically represent the same extract.
Check format, coverage, unit, language and period. Distinguish landing page from direct download; record the selected resource.
Separate record modification, dataset publication and file revision. Check the relevant licence and restrictions; public access alone does not establish reuse rights.
Link file, producer, extract and transformations. Retain checksum and parameters; identify who performed cleaning.
Reading workshop
This catalogue is entirely fictional. Four linked objects have different periods and documentary roles.
Fictional data and records for learning: none of these results comes from a live service query.
The DEMO-D12 record links a March CSV, an annual ZIP and a current API. Its description is corrected in June. The aim is to document the state published for March, rather than the latest available state.
If the table exceeds the screen, scroll horizontally; arrow keys work when the region is focused.
| Fictional object | Content or role | Possible use | Separate check |
|---|---|---|---|
| March 2026 CSV | Monthly distribution | Candidate for the March study | Check this distribution’s version, fields and licence. |
| 2025 annual ZIP | Distribution covering another period | Historical context only here | Do not add it to the CSV; read its own rights. |
| Current API | Service returning the current state | Current monitoring, subject to parameters | Retain request and response; check service coverage and rights. |
| Record modified in June | Catalogue description | Explain a metadata change | Does not establish a revision of the March CSV. |
Open each proposition to compare its diagnosis. More checks may still be needed.
Not without checking that it returns the required version. A current response may include later revisions.
Yes, as a documented candidate. Identify the exact file and check applicable rights before reuse.
No: periods and aggregation differ. Being in the same dataset does not make that sum meaningful.
Dataset: DEMO-D12 Question: state published for March 2026 Selected resource: March CSV distribution Record: description corrected in June File version: to be checked Distribution licence: not yet verified
“Not verified” marks an open action, not permission to reuse. Record date and file version remain separate fields.
In a fictional case, monthly CSV and yearly archive share a title. Choose the period and retain distribution identity; do not sum both versions.
An illustrative record’s date changes after its description is corrected. This alone does not establish revised observations.
Guided reading of an excerpt
The fictional catalogue links a dataset, an April distribution and a current service. Updating its record does not make these resources interchangeable.
Fictional exercise: identifiers, values and dates illustrate the check. This excerpt is not a live collected response.
D-42 names the overall subject. It does not identify the version actually analysed.
D-42-2026-04 announces an April distribution. Verify coverage and content, not merely its name.
API-live provides current access. Historical analysis needs a suitable snapshot or documented parameters.
dataset: D-42
distribution: D-42-2026-04
mediaType: text/csv
period: 2026-04
service: API-liveNo. The resource must match the required period and version. A current service may have revised data; an older distribution also needs checking before use.
Official references : W3C — DCAT 3
Worked exercise
Fictional example: the values and documents teach the method and do not describe a real organisation.
No new observation year is announced. You need data to answer a question about 2026.
If the table exceeds the screen, scroll horizontally; arrow keys work when the region is focused.
| Object | Time reference | Reference type |
|---|---|---|
| Catalogue record | 2026 | Record update |
| CSV | 2015–2020 | Observation period |
| Documentation | 2026 | Documentation correction |
The 2026 update does not make observations current. The CSV covers six years through 2020. For a 2026 question, find an appropriate distribution or state the gap. Corrected documentation may change interpretation without adding observations.
If a 2026 CSV exists on the same record, compare its ID, dimensions and coverage before replacing the old one.
Keep distribution ID, coverage, file date, documentation and selection reason.
Documentation reference : W3C — DCAT 3, temporal coverage · Consulted on
A record connecting metadata to the exact file and its reuse conditions.
Not necessarily. It may offer several distributions or a service. Keep the page and actual resource obtained.
No. It identifies identical bytes. Definitions, coverage and accuracy require other checks.
Specificate its role: observed period, version, publication or download. Several dates may be needed to allow the reader to understand the state used and the timeliness of the measured phenomenon.
TOOLS / REGISTERS
Link file, producer, extract and transformations. Retain checksum and parameters; identify who performed cleaning.
DOCUMENTED REFERENCE
When two links appear to describe the same data, compare their objects before counting two sources. Organise collection so each value can be traced to the representation actually consulted.
| Reference | Value or form | What it identifies |
|---|---|---|
| dcat:Dataset | dcat:distribution | Dataset. |
| dcat:Distribution | dcat:accessService | Accessible representation of the dataset. |
| dcat:DataService | dcat:endpointURL | Service providing data access. |
Record the catalogue entry, representation, any query and applicable licence together. Two formats of one dataset are not independent confirmations. A catalogue date does not date every observation.
Documentation consulted on .
DOCUMENTED REFERENCE
Europeana separates the described cultural heritage object from its web representations. This prevents treating a catalogue page as an image file or one representation’s rights as universal. Class and property names follow EDM documentation checked 22 October 2026.
| Reference | Value or form | What it identifies |
|---|---|---|
| edm:ProvidedCHO | Cultural heritage object | The supplied heritage object, including born-digital objects. Its creation date is not automatically a digitised-file date. |
| edm:WebResource | Digital representation | A web-accessible digital representation. Several resources may represent one object with their own characteristics and rights. |
| ore:Aggregation | Provider aggregation | Aggregation groups links and provider provenance around the object; not a new image file or work. |
| edm:isShownAt | Contextual page | Contextual view of the object, often a presentation page; not guaranteed reusable image bytes. |
| edm:isShownBy | Digital object | Link to the online digital object. Check media type and actual response before assuming JPEG or embedding. |
| edm:hasView | Additional view | Additional view, such as a detail or another representation. Multiple views do not establish multiple heritage objects. |
| edm:rights | Resource / aggregation | A resource can carry specific rights; documentation explains aggregation-rights fallback when representation-specific rights are absent. |
| edm:dataProvider | Contributing institution | Institution contributing data, possibly through an aggregator. Not necessarily the direct technical provider or object creator. |
Retain exact representation URL, rights, credit and modifications. Documentation describes differing rights for different-resolution versions; thumbnail rights do not automatically extend to high resolution. Open the provider record when context is incomplete.
Only after checking: it may be an HTML context page. Identify the representation, check media type and rights, then credit object and provider.
Documentation consulted on .
DOCUMENTED REFERENCE
OAI-PMH 2.0 distinguishes items, metadata formats and disseminated records. Document version 8 January 2015 defines date selection and partial-list resumption. Diagram counts are protocol examples, not measured HAL or OpenAIRE totals.
| Reference | Value or form | What it identifies |
|---|---|---|
| datestamp | Metadata creation / change / deletion | Metadata record creation, modification or deletion date, not automatically publication date. |
| from · until | Inclusive bounds | Both bounds are inclusive. Resuming at the same bound can overlap prior records; check identifiers. |
| granularity | Day / optional second | Day granularity is mandatory, optional seconds support declared by Identify. Do not invent precision. |
| resumptionToken | Exclusive argument, with verb | Resume with token and verb, without adding initial selection arguments. Tokens are opaque and may expire or become invalid. |
| completeListSize | Optional estimate | Optional complete-list size can be an estimate and change during traversal. Missing is not zero; count actual records. |
| header status="deleted" | No metadata or about | Deleted records retain a header without metadata or about. Do not replace this state with empty-field records. |
| deletedRecord | no · transient · persistent | Identify declares deletion tracking: no, transient without persistence guarantee, or persistent history. |
| Protocol example | 175 = 100 + 75 | Official example splits 175 records into 100 and 75. Final token is empty; short-page logic is not a universal OAI stopping rule. |
Keep identifier, metadataPrefix and datestamp. One deleted format does not establish deletion of every format for the item. Document boundary overlap and deduplication; absent deletion history cannot be reconstructed automatically.
No for these resumption requests: token is exclusive apart from verb. Changed selection requires a new query, tracked as a distinct traversal.
Documentation consulted on .
DOCUMENTED REFERENCE
An annual release and a correction to the same file serve different purposes. DCAT 3 describes resource versions and datasets belonging to a series explicitly; these links do not replace reading the methodology.
| Reference | Value or form | What it identifies |
|---|---|---|
| dcat:hasVersion | resource → version | Links a resource to one of its versions. Record the identifier of the version actually consulted, not only the catalogue URL. |
| dcat:previousVersion | new → previous | Links a version to its predecessor. Revisions can change older observations: compare release notes before calculating a trend. |
| dcat:inSeries | dataset → series | Places a dataset in a series. Successive annual datasets can belong to a series without being corrective versions of the same dataset. |
Reading example: releases covering 2024 and 2025 form a temporal sequence; a corrected 2024 file still covers 2024. Check the observation period, publication date, revision and distribution format separately.
Documentation consulted on .
DEEPER READING / PRACTICE
A catalog describes a dataset, but the downloaded resources may have different formats or versions. The reading must bring the descriptive file and accessible files closer.
Updated on .
Release producer, dates, resources, formats, licenses and documentation. Open the files to check fields, coverage and correspondence with the description.
A date of update of the catalog does not guarantee a new measure. A resource may be unavailable or only cover a part of the dataset described.
The catalog sheet retains metadata, chosen resource, date of download and deviations. It serves as a starting point for data assessment.